Moderate, Except for the Part Where It's Exploited: Inside CVE-2026-124194 and Outlook's NTLM Hash Leak

Microsoft’s November Patch Tuesday fixed over a thousand vulnerabilities, and the only one it says was being exploited is rated Moderate. CVE-2026-124194 is an Outlook spoofing bug with a CVSS score of 6.5. It also leaks a user’s NTLMv2 hash to a server the attacker controls, and that is worth more than the number suggests.

A Note on Sourcing

The description, scores, FAQ, credit and affected products below come from Microsoft’s Security Update Guide entry for the CVE, which I read directly. The Microsoft 365 Apps build numbers come from Microsoft’s Office security release notes. The KEV listing, the Patch Tuesday counts and the mitigation advice come from write-ups by CSO Online, ThreatDefence and Safeguard, and I say so where it matters. Two of those write-ups label their mitigation lists as their own guidance and not Microsoft’s, and I have kept that distinction below.

What the Bug Does

The weakness is CWE-356, a product UI that fails to warn the user about an unsafe action. Microsoft’s FAQ describes it like this: an unauthenticated attacker sends a crafted message file and convinces a user to open it in an affected Outlook. Opening it can make Outlook connect to an attacker-controlled network location without showing the usual external-content warning. Microsoft states the result directly: the user’s NTLMv2 hash is disclosed, and the attacker could use it to authenticate as the user.

Three details limit the bug and are worth stating plainly:

  • It is not zero-click. The user has to open the file.
  • Microsoft says the Preview Pane is not an attack vector.
  • Microsoft’s list of affected products is Microsoft 365 Apps for Enterprise and Office LTSC 2021 and 2024, in 32-bit and 64-bit. New Outlook, Outlook on the web, Outlook for Mac and the mobile apps are not on it. The “classic Outlook” label comes from the write-ups, not from Microsoft’s page.

Microsoft’s exploitability assessment is “Exploitation Detected”, with the bug not publicly disclosed at release. CISA added it to the Known Exploited Vulnerabilities catalogue on 10 November with a federal deadline of 24 November, according to the write-ups. Microsoft has not said who is exploiting it, who the targets are or how the files are delivered, and it has published no indicators of compromise. One aggregator also lists GitHub repositories claiming a proof of concept, described as untested. I have not looked at them and I am not linking them.

Why 6.5 Undersells It

Microsoft’s vector is CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:N/A:N, with the temporal metrics bringing it to 6.2. That is network, low complexity, no privileges, user interaction required, high confidentiality impact and nothing else. It is an accurate description of the flaw in isolation. The score measures the bug, and it cannot measure what the leaked value is worth afterwards.

I wrote about the opposite case recently, where a high score sat on code that is hard to reach. Here the score is modest and the exposure is a single click on a file in someone’s inbox.

What an NTLMv2 Hash Buys

An NTLMv2 response is not the password, but it is useful in two ways.

Offline cracking. The attacker guesses passwords against the captured response on their own hardware. There is no lockout and no contact with your network. A long random passphrase resists this and a weak or predictable password does not. I walked through the economics in the Password Cracking series, where NetNTLMv2 sits in the hash landscape.

Relay. The attacker passes the response to another service in real time and acts as the user there. This works against targets that do not require SMB signing, LDAP signing and channel binding, or Extended Protection for Authentication on HTTP services. It needs no cracking at all, and Impacket’s ntlmrelayx is the tool I used to show it. If your environment still accepts unsigned relays, a leaked hash from one laptop can become a session on a file server or a certificate from an internal CA.

That is the same reason on-prem Exchange is an Active Directory attack surface. One user credential is rarely the end of the story, and what matters is what else accepts it.

Which Builds Are Fixed

The fix ships as an Office update and not through Windows Update, so Outlook has to be updated and restarted. Click-to-Run applies it only while the Office apps are closed. The 10 November builds below match Microsoft’s Office security release notes for Microsoft 365 Apps, and the LTSC build numbers are the ones on the CVE page itself:

Product or channelFixed build
Current Channel 261020527.20196
Monthly Enterprise 2610 / 2609 / 260820527.20194 / 20430.20264 / 20326.20320
Semi-Annual Enterprise 2610 / 2609 / 260820527.20192 / 20430.20262 / 20326.20318
Office LTSC 202416.0.17932.21140
Office LTSC 202116.0.14334.21048

Office 2016 and 2019 are not in Microsoft’s affected list, and both are out of support. ThreatDefence advises treating any copy you still run as exposed, which is sensible when the vendor has not said either way.

Closing the Path

Patching fixes this bug. The same class of leak has come back before, so I would also close the route it uses. These steps are the write-ups’ guidance, not Microsoft’s.

Block outbound SMB at the perimeter. The leak needs the client to reach an attacker’s server on TCP 445 or 139. There is no reason for a user machine to speak SMB to an internet address. On a FortiGate, a deny rule looks like this. Replace the interface names with yours and make sure the rule sits above the policy that permits LAN to internet traffic, or it will never match.

config firewall service custom
    edit "SMB-NETBIOS-OUT"
        set tcp-portrange 139 445
    next
end
config firewall policy
    edit 0
        set name "Deny-SMB-to-Internet"
        set srcintf "lan"
        set dstintf "wan1"
        set action deny
        set srcaddr "all"
        set dstaddr "all"
        set schedule "always"
        set service "SMB-NETBIOS-OUT"
        set logtraffic all
    next
end

Logging matters as much as blocking. Any hit on that rule after the patch tells you a machine tried to authenticate to an external host.

Stop the WebDAV fallback. ThreatDefence points out that if port 445 is blocked, Windows can fall back to WebDAV over HTTP. Stopping the WebClient service on machines that do not need it closes that route.

Restrict outgoing NTLM. The Group Policy setting is “Network security: Restrict NTLM: Outgoing NTLM traffic to remote servers”. The registry value is RestrictSendingNTLMTraffic under HKLM\SYSTEM\CurrentControlSet\Control\Lsa\MSV1_0, where 1 audits and 2 denies. Audit first, because plenty of legacy things still send NTLM and you want to see them before you break them.

Put privileged accounts in Protected Users. Members cannot authenticate with NTLM, so there is no hash to leak. Test before you add anyone.

Close the relay paths. Require SMB signing, LDAP signing with channel binding, and Extended Protection for Authentication on AD CS web enrollment and Exchange. Write-ups also mention that November’s Windows updates start an NTLM signing change tracked as KB5135557. I have not verified what it covers, and it reduces relay exposure without stopping the leak itself.

What to Look For

No indicators have been published, so detection is behavioural:

  • Event 8001 in the Microsoft-Windows-NTLM/Operational log records outgoing NTLM authentication with the target server name, when outgoing-NTLM auditing is on. An external server name there is a direct sign of a leak.
  • Your firewall logs. Allowed outbound traffic on 445 or 139 means your egress rule has a gap.
  • Event 4624 with logon type 3 and NTLM from an unexpected source for an account, which can mean a cracked or relayed credential.
  • If a leak is suspected: reset the password, revoke sessions and refresh tokens, check for new certificates issued to the account, and remove the message from every mailbox.

Who Found It

Microsoft’s acknowledgements list two names: Benoit Sevens, credited as using Claude with Anthropic, and nono. That fits a pattern I have covered this year, from Cisco’s hardening advisories crediting frontier models to the judgment gap that follows when more findings arrive than anyone can validate. The credit says who reported the bug. It does not say how the attackers who were already exploiting it found it, and nobody has published that.

The Counts Don’t Agree

The headline number for this Patch Tuesday depends on where you read it. Coverage gives roughly 1,090, 1,093, 1,094 and 1,235 fixes, and at least one write-up notes the total includes CVEs Microsoft republished for third-party code. The Outlook bug is the one entry on the list that Microsoft says was already in use.

Patch Outlook first and confirm the build on every endpoint that has it. Then check whether any of your machines can still open a connection to an internet address on port 445.